>What about loops in the network?  If someone "beyond" PSUVM sends mail
>to CEARN (for example) it is easily possible that the file would go
>through the BITNET-EARNET link on the way to Europe but that a reply
>would come back through DEARN-GWUVM instead.  If the networking software
>were expecting all files from a node to come a certain direction, all
>such loops would have to be eliminated and routing would have to stay
>static.
I'd think that it should be possible to teach it that some set of nodes
have multiple paths to <reach/be reached by> them. I don't see this as
a major problem (assuming that you know the entire network topology,
which may not be reasonable).
-njg